AI and data consultant for UK charities
One question. One cited answer.
Your CRM, your accounts, your spreadsheets and your documents each hold part of the picture, and none of them talk to each other. I build private AI assistants that read across all of them at once and give you a single, checkable answer, in seconds instead of a morning of exports. Nothing leaves your control.
You'd ask things like: “Our spring appeal raised £18,240. The accounts show £17,865. Where’s the difference?” (answered from CiviCRM and Xero)
In this video, the demo reads all four of a sample charity's systems at once to pull one wetland appeal update together: its CRM (CiviCRM), its finance system (Xero), its documents (Google Docs) and its survey log (Google Sheets). Nothing in this demo left the private account it runs in. See the full walkthrough →
This is what I build for UK charities: a private AI assistant connected to the systems you already run, on data that never leaves your control. Ask it a question and it reads across whichever of your systems the answer needs, one or all of them at once, and catches what no single system can show on its own.
Your systems already hold the answers; the work is getting them out. Private AI assistants do exactly that, and three things make them different from pasting into a public tool.
Ask in plain English and they connect to whichever of your systems a question needs (your CRM, your accounts, your spreadsheets, your documents), reading them live, not from a stale export. Each one can draw on a single system or several at once, joining what it finds into one answer, even two systems that never quite agree.
Every figure they give you is a link. Click it and the real record opens: the contact in your CRM, the transaction in your accounts, the row in the spreadsheet, the paragraph in the policy. Nothing is taken on trust.
Your private AI assistants run privately, in your own AWS account in London, powered by an open weights AI model that runs inside that account rather than calling out to a US company.
They read your systems with their own limited, read only access: they can look things up, but cannot change or delete a record. Your questions, your records and their answers are never sent to ChatGPT, Gemini, Claude or any other outside service, so even a safeguarding note is safe to ask about.
Your private AI assistants feel like the AI your team already uses, only more powerful, because they read your actual data. And it all stays in your own systems.
You have probably seen AI make things up. There is a word for it, hallucinating: the model states something false as confidently as it states something true. With any AI it is a real risk, and the right first question to ask.
Here is how your private AI assistants are built to avoid it. Four habits.
1. They never do the arithmetic. Every count and total is worked out by ordinary code reading your systems. The AI's job is to phrase the answer, not to work it out.
2. They would rather say "I don't know" than guess. If the answer isn't in your systems, they tell you that, instead of filling the gap with something that sounds right. An answer that can't be checked is held back before it reaches you.
3. Every figure links to its source, and the link is built from the record itself. The assistant never writes the link, so it cannot invent one. Click it and the real record opens.
4. They are tested before they reach your team. Against data where the right answers are already known, again and again, with a single wrong answer counting as a fail.
Here's how I test AI assistants and what the testing catches →
Whatever your policy says, people on your team are probably already pasting charity data into ChatGPT, Gemini or Claude to get through the day. Every paste sends it to a US company, into logs you cannot see, and under UK law the responsibility stays with you.
The usual answer is more governance: a policy, some training, a line in the risk register. But a policy is a rule you are hoping people follow, and they break it because the public tools genuinely help them get the work done.
So the fix isn't a stricter policy. It's removing the reason anyone opens ChatGPT in the first place.
Give your team the same help in a place your data is allowed to be, and there is nothing left to reach for. You have not banned something useful; you have made the useful thing safe, so the policy stops being the only thing standing between you and a bad afternoon.
More on why this matters, and where charities are exposed today: the shadow AI problem, and why a policy won't fix it →
"He is a joy to work with as he is flexible, efficient and a great communicator. We now have a system which is far more intuitive for our visitors and 7,000+ members, which ultimately allows us to better achieve our mission of supporting the interdisciplinary research community."
Joe Yates & Steph Ray, The London School of Hygiene & Tropical Medicine
Most charities might start with one assistant, pointed at the job that's hurting, and prove it there. Some might stop at one; others might add a second, then a fifth. Here are some common starting points that you can think about for your charity, each a real build with a demo you can watch.
Ask one question that spans your CRM, finance, spreadsheets and documents (CiviCRM, Xero, Google Sheets, etc.) and catch the answers, or inconsistencies, no single system can show on its own.
See the walkthrough →Ask plain questions of your supporter and giving data in Beacon, Donorfy, CiviCRM, etc., and get answers without it leaving your control.
See how →Track retention, growth and engagement across your membership in Beacon, CiviCRM, Salesforce, etc., on data that never leaves your systems.
See how →Turn long case histories into a clear handover or report, straight from Lamplight, Charitylog, a Google Docs folder, etc., all inside your own perimeter.
See how →Ask your charity's own policies, board papers and funder contracts a question, wherever they live (Google Drive, SharePoint, Dropbox, etc.), and get the answer with its source cited.
See how →That list isn't the limit. If your system isn't named above, that's usually fine: if it has a way in, an API, a database, even a regular export, I can build a secure connector so an assistant can reach it too. Tell me what you're trying to do →
"We have worked with many developers and can confidently say that Pete is by far the best, an exceptional coder and consultant with an impressive skill set."
Kathryn Maxwell, IT Project Manager, Royal Meteorological Society
I've spent over 25 years building the data architecture under big software for UK organisations.
The platforms have changed (Perl and Python before the web was the default, Drupal and Laravel through the 2010s, private AI today) but the work hasn't. It's still about getting messy data out of legacy systems and into shape so something useful can sit on top.
Since 2009 I've done that for over 45 UK organisations. These assistants are just the new layer on top. The data plumbing underneath is still the hard part.
A few you may recognise:
The first step is always the same, and it's a small one: a short, fixed-price discovery. Over a couple of weeks I work out what your team is already doing with AI, where your data actually lives, and the one thing worth building first. You get a written report and a call to talk it through, with no obligation to go further. It's genuinely useful on its own, whether or not we end up building anything.
Here's a sample, laid out exactly as the real one is delivered.
For context: I work mainly with UK charities and non profits, with chief executives, operations and finance directors, programme leads, and the people who look after data and IT. Respectfully, I don't work with recruitment or development agencies.
Not sure it's time for that yet? Just email me, tell me who you are and what your organisation does: peter@peterbrady.co.uk
PNB Technologies Limited
55 Yew Tree Road, Ormskirk
Lancashire, L39 1NT, United Kingdom
Company Reg: 07166600
VAT Reg: GB986726850
© 2026 Peter Brady. All rights reserved.